Forgeprint

Catalog + MCP server · for any coding agent

Blueprints, experts and crews for coding agents — resolved for you.

Say what you know and what you are building. Forgeprint returns exactly one blueprint — the rule file, the skills, the MCP configuration and a setup recipe your agent runs step by step — and, when you ask, the experts and tools to work on it with.

Pre-release · 46 blueprints · 47 experts · 20 crews · 45 integrations · 9 agents

Connect it to your agent

The examples on this page follow the agent you pick. 

claude mcp add forgeprint -- npx -y forgeprint-mcp

One command. On Windows PowerShell, quote the separator: "--".

Tested end to end: an agent ran a whole recipe through it.

Browse the catalog

Every new project starts cold

You hunt down the requirements, write the rule file by hand, wire up the MCP servers, pick the plugins, and get the setup steps wrong twice before they work. Hours of it — repeated by everyone, on every project, forever.

That knowledge is not hard to write down. It is just never written down in a form another agent can use.

Without Forgeprint

  • Search for a starting point, find five, compare them by hand
  • Write AGENTS.md from memory, forget half of it
  • Wire MCP servers one at a time, from their own READMEs
  • Tell the agent how to review, test and ship — again, in every project
  • Discover the gaps only when something fails

With Forgeprint

  • Describe yourself in one sentence; the resolver asks what it still needs
  • Get one blueprint, with the reason it was chosen
  • The rule file, the skills and the MCP configuration arrive together
  • Experts bring a way of working, with checklists and sources behind it
  • The recipe is executed in CI before it ever reaches you

Three questions, four units

A blueprint answers what you are building. Two more questions come up in the same breath — how the agent should work, and with which tools — and each has its own unit. A crew packages the answers under a name.

What are you building?

Blueprint

A project package: AGENTS.md, skills, MCP configuration, and setup.md — a recipe whose every step ends in the check that proves it worked.

How should the agent work?

Expert

A way of working, not a persona. It names what it produces, the checklists it applies, and the sourced references it rests on.

With which tools?

Integration

A pinned, permission-documented install recipe for somebody else's MCP server. Forgeprint hosts no third-party code.

All of it, under one name?

Crew

Up to six experts and their integrations, named by slug. It copies nothing, and says what it is for — and what it is not for.

How the units point at each other — live from the catalog

How it works

You talk to your agent normally. Forgeprint answers in structured data, and your agent does the work. It installs nothing and runs nothing on your machine — every tool returns text.

  1. You say who you are
  2. Your agent calls resolve
  3. One blueprint comes back
  4. It asks rather than guesses
  5. You answer
  6. It fetches the package
  7. Context and recipe arrive together
  8. Your agent runs the recipe

Forgeprint returns text. It never runs anything on your machine.

What it looks like

Four conversations, one for each unit. The tool calls are shown the way your agent makes them.

YouI know C#. I want to build a multi-tenant SaaS API.

resolve { languages: ["C#"], goal: "multi-tenant SaaS API" }

Your agentOne blueprint fits: dotnet-multitenant-saas-api — chosen because you already write C#, and it covers multi-tenancy. Two things are still open: Postgres or SQL Server, and one shared database with a tenant column, or one database per tenant?

YouPostgres, one shared database.

Your agentI have the recipe: numbered steps, pinned versions, and a check after each one. Shall I start in this folder?

Ten tools, three jobs

Everything Forgeprint does is one of these calls. Each returns text; none of them runs anything.

Start a project

resolve
Your profile in; either the questions to ask you, or one blueprint with the reason.
search_blueprints
Stack, language, type or text in; matching blueprints with their scores out.
compare_blueprints
Two to four slugs in; a pros-and-cons table built from their own overview files.
get_blueprint
The package, with the recipe's option blocks resolved for your choices.

Work with the right people

recommend_experts
A task in; at most one crew or three experts, each with a reason.
get_expert
SKILL.md, checklists and references — and the paths your agent reads them from.
get_crew
Every member in full, and the install command for each integration, for your agent.
get_integration
Pinned upstream, install command, and the secrets it needs with what they reach.

Grow the catalog

validate_blueprint
A blueprint folder in; schema errors and a duplicate report out.
request_blueprint
When nothing fits: an issue payload. The catalog grows by demand.

Ask resolve for an expert and it does not improvise one: it tells your agent which tool to call instead.

Written once. Rendered for nine agents.

Content lives in the catalog once, as AGENTS.md and SKILL.md. Each agent's own file is generated from it — a hand-written copy per agent is refused. What every agent reads, and the day its documentation was last checked, is on record.

Written once

AGENTS.md · SKILL.md

Rendered per agent

forgeprint render --agent

What each agent gets

  • Claude CodeCLAUDE.md
  • Codex CLIAGENTS.md
  • GitHub Copilot CLI.github/copilot-instructions.md
  • Cursor.cursor/rules/{slug}.mdc
  • Gemini CLIGEMINI.md
  • Windsurf.windsurf/rules/{slug}.md
  • Cline.clinerules/{slug}.md
  • OpenCodeAGENTS.md
  • Kiro.kiro/steering/{slug}.md

No MCP client?

The same files, written to disk from a clone of the repository — two ways in, one answer out. Skills also install through tools you may already use. 

pnpm forgeprint get dotnet-web-api --agent claude-code
npx skills add forgeprint/forgeprint
gh skill install forgeprint/forgeprint blueprint-author

What makes it different

A resolver, not a list

One answer, weighted so a language you already know beats everything else.

A recipe, not a description

setup.md is a script: pinned versions, each step ending in the check that proves it worked.

A gate, not a junk drawer

Schema validation, a real setup run and a duplicate report on every pull request.

Standards, not another format

AGENTS.md, SKILL.md and marketplace.json as they already exist.

Composition, not copies

A crew names experts and integrations by slug. At most six members, and a line saying what it is not for.

Any agent, on the record

Nine agents registered from their vendors' own docs, each with a date. Tested means tested.

The catalog

Everything in it right now, read live from the published index. Every blueprint's recipe runs in CI on every push, in every option combination.

Full catalog with search Catalog entries stay in English. They are written once, in one language, and your agent presents them in yours.

Who it is for

People who use it

Anyone starting a project with an agent

You do not need to know what an MCP server is.

Teams who keep rewriting the same setup

A blueprint is that decision, written down once.

People who want the agent to work like a senior

An expert brings the checklists and the sources, not a costume.

People who build it

Anyone with a project that works

The most valuable blueprint comes from something real.

Anyone with a way of working worth writing down

An expert is a role, its checklists and its references.

People who just want one to exist

File a request. The catalog grows by demand.

How to contribute

One folder is one unit — a blueprint, an expert, a crew or an integration — and that folder is the contribution. Everything the pipeline does is a command you can run on your own machine; hosted CI is a convenience here, never a dependency.

Get it running

git clone https://github.com/forgeprint/forgeprint.git
cd forgeprint
pnpm install
pnpm run build
pnpm forgeprint validate

How the repository is laid out

blueprints/<slug>/
What you build: manifest, AGENTS.md, overview, setup.md
experts/<slug>/
How the agent works: SKILL.md, checklists, references
crews/<slug>/
Named packages of experts and integrations
integrations/<slug>/
Pinned install recipes for third-party MCP servers
schema/
Manifest schemas, the taxonomy, and agents.yaml
packages/
The CLI, the MCP server and the catalog pages
docs/decisions/
Every architectural decision, with the reasoning kept

The contribution flow

  1. Check whether it already exists

    One unit per combination: one blueprint per stack and requirements, one expert per role, domain and seniority.

  2. Write it from something that works

    The blueprint-author skill turns a working project into a blueprint.

  3. Pass the gate locally

    validate, lint-setup, similarity, render-check and a real setup run.

  4. Open the pull request

    Say what is closest and why yours is not a change to it.

Questions people actually ask

Does Forgeprint run anything on my machine?

No. Every tool returns text. Your agent is what reads the recipe and runs the commands, and you can read every step before it does.

Is it only for Claude Code?

No. Nine agents are registered, each traced to the vendor's own documentation with the date it was read. Claude Code is the first that has run a whole recipe end to end; the others are marked as documented, not tested, until somebody runs one.

What is an expert — a persona?

No. An expert is a way of working: what it produces, the checklists it applies, and references with versions and the date they were last checked.